The idea is that bots are inflexible to deviations from accepted norms and can't actually "see" rendered browser content. So if your generic 404, 403 error pages return a 200 status instead, with invisible links to other non accessible pages. The bots will follow the links but real users will not, trapping them in a kind of isolated labyrinth of recursive links (the urls should be slightly different though). It's basically how a lobster trap works if you want a visual metaphor.

The important part here is to do this chaotically. The worst sites to scrape are buggy ones. You are, in essence, deliberately following bad practices in a way real users wouldn't notice but would still influence bots.